Integrating RNA expression and visual features for immune infiltrate prediction

Area of Science:

• Computational biology
• Cancer research
• Immunology

Background:

• Patient response to cancer immunotherapy is influenced by tumor genomics and microenvironment.
• Current immunotherapies benefit only a subset of patients, highlighting a need for better predictive tools.
• An inflamed tumor microenvironment and high tumor-infiltrating immune cells correlate with better immunotherapy response.

Purpose of the Study:

• To develop a novel computational framework, NEXT (Neural-based models for integrating gene EXpression and visual Texture features), for accurate characterization of the tumor-immune microenvironment.
• To integrate RNA-sequencing (RNA-seq) data with digital pathology images for a comprehensive analysis of individual patient tumors.
• To improve the prediction accuracy of immune cell infiltration in solid tumors.

Main Methods:

• Utilized RNA-seq data and digital pathology images from patient tumors.
• Developed and applied the NEXT framework, a neural network model, to predict immune infiltrates.
• Validated NEXT predictions against expert pathology review across four cancer types.

Main Results:

• Integration of imaging features significantly improved the prediction of immune infiltrates compared to gene expression data alone.
• The enhancement in prediction accuracy was particularly notable for B cells and CD8 T cells.
• The NEXT framework demonstrated reliable and accurate prediction of immune composition in individual patient tumors.

Conclusions:

• The NEXT framework effectively integrates multi-modal data (RNA-seq and imaging) for robust analysis of the tumor immune microenvironment.
• This integrated approach offers a more accurate prediction of immune cell composition, crucial for understanding immunotherapy response.
• The findings support the utility of combining genomic and imaging data in clinical settings to personalize cancer treatment strategies.